Please place your comments or questions on ch. 2 pp. 30-115 here by the end of the day Sunday, Sept. 16th.
10 thoughts on “Response to McCarthy 2008: ch. 2a, due 9/16”
Megan
I guess I shouldn’t be surprised from a book of this title, but I was really impressed with this chapter’s useful and practical introduction to the practice of OT. John did a great job of addressing mistakes and problems that frequently arise when doing an analysis.
As one becomes more comfortable with OT, it becomes tempting to complete analyses more quickly and to rely more heavily on intuition and somewhat less on solid ranking arguments. Because of this, even someone reasonably experienced in OT could improve their analyses using this chapter. In fact, it might be even more useful for people relying heavily on intuition, since many of the mistakes addressed are cases where OT behaves counterintuitively – for example, the ranking of a tie-breaking constraint with respect to constraints that don’t distinguish between candidates. Even in the simple problem set from last week (which I admittedly rushed through), I made a few mistakes he addresses in this chapter – I should’ve read this first!
I found McCarthy’s discussion of inputs in this chapter particularly useful.
One important idea he develops is that, assuming the crosslinguistic universality of CON, forms that do not alternate can be as informative as those that do. Those markedness constraints that are violated by a language are revealed, interestingly, when nothing happens, i.e. from a faithful input-output mapping.
Some other points I found useful:
– that an analysis may be “helped along” by way of convenient regularities in the input
– that if an item has a restricted distribution then faithfulness to it must be ranked below some markedness constraint that controls the distribution
– the distinction between inputs as a theoretical construct and underlying representations as a construct of learners (and that for a given learner the latter is a subset of the former)
– that since rankings are not specific to particular contexts and are global properties of the grammar, results about ranking in one context are applicable in other contexts
The insights and generalizations given in this chapter—particularly those in the section on inputs—are indispensable in both developing an analysis in OT and fixing it when things go awry.
This chapter has been revealing for me, finally I have seen how OT works exactly. One of the things that have attracted my attention is the necessity of going beyond what you see at first sight. This may be obvious given that we are talking about theorizing and, in the end, that’s what theorizing is about… but maybe because of the fact that OT presents the data in tableaux, it can be easy for the non-specialist to think that OT amounts to “find some data and order it”.
For example, the third condition on valid ranking arguments is one of those cases in which it is important to go further away in our analysis in order to find out possible arguments that would undermine the alleged ranking (recall, Const1>>Const2 is legitimate only if there is no Const3 that could be doing the same work as Const1). If you ignore (or overlook) that there is such a Const3 you may be constructing an illegitimate ranking. Other examples of this need to go further away in the analysis include the need to test the potential effects of a new constraint on all existing claims about particular rankings (that is, considering ranking arguments in the context of the full system), checking pairs of unranked constraints, taking care not to overlook loser candidates that may tie/beat the winner, considering -sometimes- inputs that map to faithful outputs, etc. In fact, as Megan puts it, OT may be counterintuitive in some respects, and working out systematically this kind of phenomena is the only way we have to construct a satisfactory analysis.
I found this chapter (Ch.2: 30-115) in McCarthy’s book very useful and reader-friendly. Specifically, this ‘step-by-step’ introduction enables readers to apply OT to natural linguistic data in a professional and optimality-theoretic way. Furthermore, the methodological procedure (linguistics data⇒descriptive generalization⇒theoretical analysis) introduced in this chapter also works in syntax and semantics as a guiding line to conduct own research and develop theoretical analysis.
And interesting similarities between phonology (especially in the OT framework) and syntax are described throughout this chapter, as well as Ch.1 of this textbook (cf. my previous comment). For example, blocking effects (*Cunsyll/*V#/MAX in Yawelmani vs. Dative Subject/Nominative Object in Icelandic), stringency relations (IDENT/IDENTOnset vs. T-Gov/T-Lex-Gov), Grimshaw’s application of OT to syntactic do-support (Op-Spec/Ob-Hd/Stay/Full-Int), repair strategy (Reinhart 2006), Full Interpretation FI (Chomsky 1995), etc… Apparently, theoretical assumptions behind the OT framework look different from other linguistic theories (e.g. violable constraints, ranking), but I realize that fundamental insights among them are the same and all of them are trying to capture universals and variations in natural languages.
Finally, I have two clarification questions:
(i) When do we HAVE TO use “Hasse diagrams†in ranking arguments? I understood the necessity of violation tableau, comparative tableau, and combination tableau, though.
(ii) How can we delimit the number of candidate sets/output representations from the same input, because the GEN component can generate almost infinite candidates with optional (universal?) rules but some candidates are useless to determine constraint rankings? In other words, I’m wondering whether we can select a particular candidate set beforehand in a priori way. I think that this will be a difficult but important issue when actually we attempt to carry out our own OT research.
Like the others who have posted before me, I found this chapter particularly insightful into how OT works in practice and how to proceed through an OT analysis.
However, I found the section on OT Syntax difficult to accept because the types of constraints feel intuitively different, e.g. less generic, than the types of constraints for OT Phonology. The syntactic constraints require stipulations and miss fundamental generalisations about syntax.
For example, in (66), the constraint Op-Spec, which is satisfied by an operator occupying a specifier position, is claimed to account for wh-movement. What about when the wh-word is the subject and already in spec-vP? What about wh-words in unaccusatives and passives which could just raise to spec-vP? Both of these scenarios would satisfy Op-Spec yet produce ungrammatical sentences. If you were to stipulate that the operator must raise to spec-CP, you have to explain why. Maybe you propose a Satisfy-Features constraint wherein the C head bears features that wh-movement checks (like in Minimalism). Of course, if there is no C in the input (which the chapter says consists of only an argument projection), where would these features come from? If GEN can also add features to phrase heads, the candidate set becomes significantly larger.
(I can think of other problems, but this one comes right from the chapter.)
I wonder if this problem in some capacity extends to phonology as well, e.g. triggering effects?
This chapter was useful in that it was a basic introduction on what to do when making an OT analysis. McCarthy pointed out things that should be obvious, but are often done regardless, like making an assumption about the ranking of constraints without first justifying those rankings with accompanying arguments. However, the section I found the most beneficial was the section on tableaux.
Before reading this chapter I didn’t understand the purpose of the W and L notations, because I felt that violation tableaux accomplished the same thing with more detail. McCarthy also mentioned that comparison tableaux are difficult to make without first consulting the violation tableaux. For this same reason, I was under the opinion that the W/L notation was not very useful. However; he pointed out that the combination tableaux which use both violation marks, as well as the W’s and L’s from the comparison tableaux would be an ideal combination. I’ve come to agree with his perspective, because these tableaux offer a quick organization of the constraint hierarchy.
Overall, I think this was a good introduction to OT analysis.
I also like that he showed that these methods can also be used outside of phonological research (applied to syntax).
This is the chapter gave me some insight as to why OT has its proponents, but also why it has so many skeptics. It’s clearly an unconventional framework that requires complex analysis every step along the way and what’s more – constant revisions. That being said, there are few points of the chapter that stood out to me, some technical and some not.
I wonder if OT is popular among syntacticians, as it does away with the traditional tree-based approach and one can no longer describe language typology in terms of what sort of construction and movement is permitted. Although OT was a big leap for a field of phonology it seems an even bigger leap for syntax.
Tables (28) and (29) present a ranking argument for MAX v. IDENT(long) that McCarthy says relies on a specific definition of IDENT(long). The crucial decision to be made is whether or not IDENT(long) assigns a violation mark to non-literal vowel shortening (as seen in output [tax]). While I understand that this decision is crucial, I have two worries about these tableaux. One is that even though in (28) MAX is indifferent to both outputs, we have seen a similar circumstance in a past homework where the property of minimal violation has selected one candidate over the other in the case where all candidates were assigned violation marks, some just more than others. Drawing a parallel, I do not see why MAX does not select output [taxak?] therefore rendering the outputs’ performance in IDENT(long) irrelevant. Tableau (29) confuses me especially, as while it demonstrates the ranking under the assumption that it is MAX that selects the winner, it does not show the relevance of the definition of IDENT(long).
In tableau (42) McCarthy talks about a common misconception when it comes to unranked constraints. The tableau has me deeply puzzled, as I for one cannot understand why *C unsyll and *V# are unranked with respect to one another, as previous tableaux (23) (24) (33) (34) show that they are. It that case, in tableau (42) *C unsyll would make the decision in favor of the winner. Yet even supposing that these two constraints are unranked, I do not see how they would together act as neutral towards outputs [xat.k?a] and [xat.k?] instead of eliminating both. After all, they do rank aboe MAX.
Also, looking at the examination of Yoruba and Madurese on pages 91-93, I couldn’t help but wonder after the lack of a fairly obvious constraint – one where the nasalization of a consonant and the following vowel should match up, let’s call it NAccord. In this case, the interpretation of the data would be quite different, with IDENT[nasal] ranking high in Yoruba, and low in Madurese, and NAccord- the opposite.
The examination of the Full-Int constraint on page 98 got me thinking about Russian, where an unstressed do is an impossibility. Finding a concrete example from my native tongue gave me a hope for maybe one day having the confidence to apply the framework of OT to syntax.
I agree with my peers when they say that this chapter is revealing and especially useful, not only for beginners like me, but also for linguists that have been researching under the frame of OT. In this chapter you can find the guidelines on how to proceed when analyzing data from the perspective of OT and also some interesting tips and warnings based on previous works. Being told that experienced phonologists have made mistakes when analyzing their data encourages beginners not to despair when facing problems, because we can all go through it. With regards to this idea, one of the sections that I find particularly interesting and helpful is the one that talks about the limits of ranking arguments, because it gives some fundamental cues about how prevent errors that would make the analysts suffer from frustration. First of all, we have to identify the unviolated constraints in order to reduce the number of problematic candidates, since there is no point on ranking constraints that don’t conflict. Constraints in a stringency relation cannot be ranked with respect to one another either, because there is no conflict between them. McCarthy explains this relation using an analogy that refers to criminal law, which made me think in more general terms and reminded me of the semantic notions of hypernym (A term that has a more comprehensive or more general semantic scope than an other) and hyponym (An entry in a thesaurus that has a smaller semantic scope than the headword itself). Consequently, in the same way in which every oak is also a tree but not every tree is an oak, every robbery is a larceny but not all the larcenies are robberies, a violation of a Const1 can also be considered a violation of a Const2 if we say that Const2 is more stringent than Const1, because the violation of Const2 include the violations made to Const1. The last scenario in which no ranking of constraints can be undertaken would be when there is a tie between a losing candidate and the winner. In this case, McCarthy says that we have to create a tie breaking-constraint which is also unrankable. However, the example he gives is too abstract, since he doesn’t consider real data. I think that is very difficult to come up with a new constraint that won’t affect the entire analysis of the language and for this reason, I think that an example based on real data would be more helpful.
I’ve gone over this section to give an example of the good job that I think McCarthy has done in order to explain how to proceed in the Optimality Theory framework. From now on, I will definitely have this chapter within reach when I undertake the analysis of some data and I hope not to succumb to the temptation to follow my intuition but to follow his tips instead.
I have to admit that for someone who has never done any research on an OT basis before, like me, this chapter could prove to be very instructively useful. I believe that is the point of this chapter, to help identify the mistakes one could possibly make in an OT analysis (e.g. ranking certain constraints without evidentiary support from linguistic data), and extend the framework of OT to other disciplines of Linguistics.
If following the methodology in Chapter 2, one could possibly make an ligitimate analysis in OT. But my question is, even if we pay attention to every detail concerning the correct ways of doing OT, wouldn’t there just be some situations we would be inevitably faced with regardless of how strictly we follow the instructions? For example, OT only provides the formalism of how an optimal candidate is chosen under a specific ranking of constraints. It does not provide any characterization of constraints. If so, how can we be sure if the constraints that are in effect make an legitimate analysis? One of the three criteria that make the analysis legitimate is that the ranking of two constraints is only proper when there is no third constraint that can do the job of the higher ranking constraint of the two. Suppose there is no exhaustive list of constraints in natural languages (at least no one can say for sure that he knows all the constraints in natural languages), there is no way we can know if the ranking in our analysis is proper because there could always be some constraint that we do not know of that is doing the job of a higher ranking constraint in our analysis and leaving our ranking problematic.
At the end of the day, we still have to rely on an intuitive judgment to tell if we are doing OT the right way or just simply ignore the incontrollable factors that could pose a threat to the legitimacy of our analysis. If only human language were simple enough to be analyzed with just a couple of constraints. Knowing what to do and how to do it right in OT is a start, but it seems that we cannot avoid being skeptical about the results of OT analysis. Maybe only when we gain more experiences in doing OT will we have more confidence to say that we are doing OT the right way.
I’m glad to hear people are finding the reading useful!
Here are a few comments on other things that came up above that we haven’t had a chance to talk about in class (e.g. we’ve talked a bit about Hasse diagrams, and will talk more):
1. I think Ethan’s comment raises the question that a lot of people working on syntax in OT wrestle with – what is an input? In phonology, P&S 1993 and most subsequent researchers have simply assumed that the input is the Underlying Representation of earlier generative work. A good way of thinking about the input, that came up in a conversation I had with Brian Smith last week (who could give you references), is that the input defines the set of representations that are competing with one another. In other words, the input doesn’t necessarily have to correspond to a derivational level from some theory, and the output may include multiple levels of representation, evaluated all at once. I can give you examples from phonology if you’d like, but this is really something for out-of-class discussion, since it’s complex, and will take us far from where we are currently.
2. That all being said, I think it’s important to emphasize that OT syntax can in general operate with the same types of representations, constraints, and even movement operations as other syntactic frameworks. We’ll see an example in our discussion of learning Tuesday (Intro to Learning handout). Sacha – I’d be interested to know why you seem to think otherwise, but again, this seems like something more for an office discussion.
3. In response to another of Sacha’s comments, Tableau (42) assumes that when 2 constraints are unranked, they act like a single constraint. This is sometimes called “pooled” violations, and it’s in fact how Tesar and Smolensky treat ties in their learning work. Boersma has a recent LI squib showing how it can lead to problems, though RCD, which we are now studying, works fine with this assumption, even as a learning algorithm for a version of OT with a total order on the constraints.
4. Cova’s call for a real example is a good one, but I need to give this some more thought…
5. Hsin-Lun’s comment is a commonly voiced one, and it might be useful to recall the terminological distinction I suggested in class: OT is a framework, and OT with a set of constraints is a theory. It’s the job of every phonologist working with OT, including you, to develop theories of individual languages and typology – OT itself doesn’t provide one!
I guess I shouldn’t be surprised from a book of this title, but I was really impressed with this chapter’s useful and practical introduction to the practice of OT. John did a great job of addressing mistakes and problems that frequently arise when doing an analysis.
As one becomes more comfortable with OT, it becomes tempting to complete analyses more quickly and to rely more heavily on intuition and somewhat less on solid ranking arguments. Because of this, even someone reasonably experienced in OT could improve their analyses using this chapter. In fact, it might be even more useful for people relying heavily on intuition, since many of the mistakes addressed are cases where OT behaves counterintuitively – for example, the ranking of a tie-breaking constraint with respect to constraints that don’t distinguish between candidates. Even in the simple problem set from last week (which I admittedly rushed through), I made a few mistakes he addresses in this chapter – I should’ve read this first!
I found McCarthy’s discussion of inputs in this chapter particularly useful.
One important idea he develops is that, assuming the crosslinguistic universality of CON, forms that do not alternate can be as informative as those that do. Those markedness constraints that are violated by a language are revealed, interestingly, when nothing happens, i.e. from a faithful input-output mapping.
Some other points I found useful:
– that an analysis may be “helped along” by way of convenient regularities in the input
– that if an item has a restricted distribution then faithfulness to it must be ranked below some markedness constraint that controls the distribution
– the distinction between inputs as a theoretical construct and underlying representations as a construct of learners (and that for a given learner the latter is a subset of the former)
– that since rankings are not specific to particular contexts and are global properties of the grammar, results about ranking in one context are applicable in other contexts
The insights and generalizations given in this chapter—particularly those in the section on inputs—are indispensable in both developing an analysis in OT and fixing it when things go awry.
This chapter has been revealing for me, finally I have seen how OT works exactly. One of the things that have attracted my attention is the necessity of going beyond what you see at first sight. This may be obvious given that we are talking about theorizing and, in the end, that’s what theorizing is about… but maybe because of the fact that OT presents the data in tableaux, it can be easy for the non-specialist to think that OT amounts to “find some data and order it”.
For example, the third condition on valid ranking arguments is one of those cases in which it is important to go further away in our analysis in order to find out possible arguments that would undermine the alleged ranking (recall, Const1>>Const2 is legitimate only if there is no Const3 that could be doing the same work as Const1). If you ignore (or overlook) that there is such a Const3 you may be constructing an illegitimate ranking. Other examples of this need to go further away in the analysis include the need to test the potential effects of a new constraint on all existing claims about particular rankings (that is, considering ranking arguments in the context of the full system), checking pairs of unranked constraints, taking care not to overlook loser candidates that may tie/beat the winner, considering -sometimes- inputs that map to faithful outputs, etc. In fact, as Megan puts it, OT may be counterintuitive in some respects, and working out systematically this kind of phenomena is the only way we have to construct a satisfactory analysis.
I found this chapter (Ch.2: 30-115) in McCarthy’s book very useful and reader-friendly. Specifically, this ‘step-by-step’ introduction enables readers to apply OT to natural linguistic data in a professional and optimality-theoretic way. Furthermore, the methodological procedure (linguistics data⇒descriptive generalization⇒theoretical analysis) introduced in this chapter also works in syntax and semantics as a guiding line to conduct own research and develop theoretical analysis.
And interesting similarities between phonology (especially in the OT framework) and syntax are described throughout this chapter, as well as Ch.1 of this textbook (cf. my previous comment). For example, blocking effects (*Cunsyll/*V#/MAX in Yawelmani vs. Dative Subject/Nominative Object in Icelandic), stringency relations (IDENT/IDENTOnset vs. T-Gov/T-Lex-Gov), Grimshaw’s application of OT to syntactic do-support (Op-Spec/Ob-Hd/Stay/Full-Int), repair strategy (Reinhart 2006), Full Interpretation FI (Chomsky 1995), etc… Apparently, theoretical assumptions behind the OT framework look different from other linguistic theories (e.g. violable constraints, ranking), but I realize that fundamental insights among them are the same and all of them are trying to capture universals and variations in natural languages.
Finally, I have two clarification questions:
(i) When do we HAVE TO use “Hasse diagrams†in ranking arguments? I understood the necessity of violation tableau, comparative tableau, and combination tableau, though.
(ii) How can we delimit the number of candidate sets/output representations from the same input, because the GEN component can generate almost infinite candidates with optional (universal?) rules but some candidates are useless to determine constraint rankings? In other words, I’m wondering whether we can select a particular candidate set beforehand in a priori way. I think that this will be a difficult but important issue when actually we attempt to carry out our own OT research.
Like the others who have posted before me, I found this chapter particularly insightful into how OT works in practice and how to proceed through an OT analysis.
However, I found the section on OT Syntax difficult to accept because the types of constraints feel intuitively different, e.g. less generic, than the types of constraints for OT Phonology. The syntactic constraints require stipulations and miss fundamental generalisations about syntax.
For example, in (66), the constraint Op-Spec, which is satisfied by an operator occupying a specifier position, is claimed to account for wh-movement. What about when the wh-word is the subject and already in spec-vP? What about wh-words in unaccusatives and passives which could just raise to spec-vP? Both of these scenarios would satisfy Op-Spec yet produce ungrammatical sentences. If you were to stipulate that the operator must raise to spec-CP, you have to explain why. Maybe you propose a Satisfy-Features constraint wherein the C head bears features that wh-movement checks (like in Minimalism). Of course, if there is no C in the input (which the chapter says consists of only an argument projection), where would these features come from? If GEN can also add features to phrase heads, the candidate set becomes significantly larger.
(I can think of other problems, but this one comes right from the chapter.)
I wonder if this problem in some capacity extends to phonology as well, e.g. triggering effects?
This chapter was useful in that it was a basic introduction on what to do when making an OT analysis. McCarthy pointed out things that should be obvious, but are often done regardless, like making an assumption about the ranking of constraints without first justifying those rankings with accompanying arguments. However, the section I found the most beneficial was the section on tableaux.
Before reading this chapter I didn’t understand the purpose of the W and L notations, because I felt that violation tableaux accomplished the same thing with more detail. McCarthy also mentioned that comparison tableaux are difficult to make without first consulting the violation tableaux. For this same reason, I was under the opinion that the W/L notation was not very useful. However; he pointed out that the combination tableaux which use both violation marks, as well as the W’s and L’s from the comparison tableaux would be an ideal combination. I’ve come to agree with his perspective, because these tableaux offer a quick organization of the constraint hierarchy.
Overall, I think this was a good introduction to OT analysis.
I also like that he showed that these methods can also be used outside of phonological research (applied to syntax).
This is the chapter gave me some insight as to why OT has its proponents, but also why it has so many skeptics. It’s clearly an unconventional framework that requires complex analysis every step along the way and what’s more – constant revisions. That being said, there are few points of the chapter that stood out to me, some technical and some not.
I wonder if OT is popular among syntacticians, as it does away with the traditional tree-based approach and one can no longer describe language typology in terms of what sort of construction and movement is permitted. Although OT was a big leap for a field of phonology it seems an even bigger leap for syntax.
Tables (28) and (29) present a ranking argument for MAX v. IDENT(long) that McCarthy says relies on a specific definition of IDENT(long). The crucial decision to be made is whether or not IDENT(long) assigns a violation mark to non-literal vowel shortening (as seen in output [tax]). While I understand that this decision is crucial, I have two worries about these tableaux. One is that even though in (28) MAX is indifferent to both outputs, we have seen a similar circumstance in a past homework where the property of minimal violation has selected one candidate over the other in the case where all candidates were assigned violation marks, some just more than others. Drawing a parallel, I do not see why MAX does not select output [taxak?] therefore rendering the outputs’ performance in IDENT(long) irrelevant. Tableau (29) confuses me especially, as while it demonstrates the ranking under the assumption that it is MAX that selects the winner, it does not show the relevance of the definition of IDENT(long).
In tableau (42) McCarthy talks about a common misconception when it comes to unranked constraints. The tableau has me deeply puzzled, as I for one cannot understand why *C unsyll and *V# are unranked with respect to one another, as previous tableaux (23) (24) (33) (34) show that they are. It that case, in tableau (42) *C unsyll would make the decision in favor of the winner. Yet even supposing that these two constraints are unranked, I do not see how they would together act as neutral towards outputs [xat.k?a] and [xat.k?] instead of eliminating both. After all, they do rank aboe MAX.
Also, looking at the examination of Yoruba and Madurese on pages 91-93, I couldn’t help but wonder after the lack of a fairly obvious constraint – one where the nasalization of a consonant and the following vowel should match up, let’s call it NAccord. In this case, the interpretation of the data would be quite different, with IDENT[nasal] ranking high in Yoruba, and low in Madurese, and NAccord- the opposite.
The examination of the Full-Int constraint on page 98 got me thinking about Russian, where an unstressed do is an impossibility. Finding a concrete example from my native tongue gave me a hope for maybe one day having the confidence to apply the framework of OT to syntax.
I agree with my peers when they say that this chapter is revealing and especially useful, not only for beginners like me, but also for linguists that have been researching under the frame of OT. In this chapter you can find the guidelines on how to proceed when analyzing data from the perspective of OT and also some interesting tips and warnings based on previous works. Being told that experienced phonologists have made mistakes when analyzing their data encourages beginners not to despair when facing problems, because we can all go through it. With regards to this idea, one of the sections that I find particularly interesting and helpful is the one that talks about the limits of ranking arguments, because it gives some fundamental cues about how prevent errors that would make the analysts suffer from frustration. First of all, we have to identify the unviolated constraints in order to reduce the number of problematic candidates, since there is no point on ranking constraints that don’t conflict. Constraints in a stringency relation cannot be ranked with respect to one another either, because there is no conflict between them. McCarthy explains this relation using an analogy that refers to criminal law, which made me think in more general terms and reminded me of the semantic notions of hypernym (A term that has a more comprehensive or more general semantic scope than an other) and hyponym (An entry in a thesaurus that has a smaller semantic scope than the headword itself). Consequently, in the same way in which every oak is also a tree but not every tree is an oak, every robbery is a larceny but not all the larcenies are robberies, a violation of a Const1 can also be considered a violation of a Const2 if we say that Const2 is more stringent than Const1, because the violation of Const2 include the violations made to Const1. The last scenario in which no ranking of constraints can be undertaken would be when there is a tie between a losing candidate and the winner. In this case, McCarthy says that we have to create a tie breaking-constraint which is also unrankable. However, the example he gives is too abstract, since he doesn’t consider real data. I think that is very difficult to come up with a new constraint that won’t affect the entire analysis of the language and for this reason, I think that an example based on real data would be more helpful.
I’ve gone over this section to give an example of the good job that I think McCarthy has done in order to explain how to proceed in the Optimality Theory framework. From now on, I will definitely have this chapter within reach when I undertake the analysis of some data and I hope not to succumb to the temptation to follow my intuition but to follow his tips instead.
I have to admit that for someone who has never done any research on an OT basis before, like me, this chapter could prove to be very instructively useful. I believe that is the point of this chapter, to help identify the mistakes one could possibly make in an OT analysis (e.g. ranking certain constraints without evidentiary support from linguistic data), and extend the framework of OT to other disciplines of Linguistics.
If following the methodology in Chapter 2, one could possibly make an ligitimate analysis in OT. But my question is, even if we pay attention to every detail concerning the correct ways of doing OT, wouldn’t there just be some situations we would be inevitably faced with regardless of how strictly we follow the instructions? For example, OT only provides the formalism of how an optimal candidate is chosen under a specific ranking of constraints. It does not provide any characterization of constraints. If so, how can we be sure if the constraints that are in effect make an legitimate analysis? One of the three criteria that make the analysis legitimate is that the ranking of two constraints is only proper when there is no third constraint that can do the job of the higher ranking constraint of the two. Suppose there is no exhaustive list of constraints in natural languages (at least no one can say for sure that he knows all the constraints in natural languages), there is no way we can know if the ranking in our analysis is proper because there could always be some constraint that we do not know of that is doing the job of a higher ranking constraint in our analysis and leaving our ranking problematic.
At the end of the day, we still have to rely on an intuitive judgment to tell if we are doing OT the right way or just simply ignore the incontrollable factors that could pose a threat to the legitimacy of our analysis. If only human language were simple enough to be analyzed with just a couple of constraints. Knowing what to do and how to do it right in OT is a start, but it seems that we cannot avoid being skeptical about the results of OT analysis. Maybe only when we gain more experiences in doing OT will we have more confidence to say that we are doing OT the right way.
I’m glad to hear people are finding the reading useful!
Here are a few comments on other things that came up above that we haven’t had a chance to talk about in class (e.g. we’ve talked a bit about Hasse diagrams, and will talk more):
1. I think Ethan’s comment raises the question that a lot of people working on syntax in OT wrestle with – what is an input? In phonology, P&S 1993 and most subsequent researchers have simply assumed that the input is the Underlying Representation of earlier generative work. A good way of thinking about the input, that came up in a conversation I had with Brian Smith last week (who could give you references), is that the input defines the set of representations that are competing with one another. In other words, the input doesn’t necessarily have to correspond to a derivational level from some theory, and the output may include multiple levels of representation, evaluated all at once. I can give you examples from phonology if you’d like, but this is really something for out-of-class discussion, since it’s complex, and will take us far from where we are currently.
2. That all being said, I think it’s important to emphasize that OT syntax can in general operate with the same types of representations, constraints, and even movement operations as other syntactic frameworks. We’ll see an example in our discussion of learning Tuesday (Intro to Learning handout). Sacha – I’d be interested to know why you seem to think otherwise, but again, this seems like something more for an office discussion.
3. In response to another of Sacha’s comments, Tableau (42) assumes that when 2 constraints are unranked, they act like a single constraint. This is sometimes called “pooled” violations, and it’s in fact how Tesar and Smolensky treat ties in their learning work. Boersma has a recent LI squib showing how it can lead to problems, though RCD, which we are now studying, works fine with this assumption, even as a learning algorithm for a version of OT with a total order on the constraints.
4. Cova’s call for a real example is a good one, but I need to give this some more thought…
5. Hsin-Lun’s comment is a commonly voiced one, and it might be useful to recall the terminological distinction I suggested in class: OT is a framework, and OT with a set of constraints is a theory. It’s the job of every phonologist working with OT, including you, to develop theories of individual languages and typology – OT itself doesn’t provide one!